Automobile axle fault diagnosis method and system

Abstract

The invention relates to the technical field of fault diagnosis, and particularly discloses an automobile axle fault diagnosis method and system, and the method comprises the steps: collecting an axle abnormal sound signal in a driving process through a sound sensor, carrying out the fluctuation analysis of the axle abnormal sound signal through a data processing algorithm based on deep learning, and carrying out the fault diagnosis of an automobile axle. And meanwhile, the understanding of the sound characteristics of the abnormal sound of the axle is enhanced by combining abnormal sound description information input by a user, and fine-grained cross-modal response collaborative reasoning is performed on the fluctuation mode of the sound signal of the abnormal sound of the axle and the abnormal sound description information, so that potential axle fault characteristics are mined, and intelligent diagnosis of fault parts of the automobile axle is realized. Through the mode, the dependence of a traditional automobile axle fault diagnosis method on artificial experience can be effectively overcome, meanwhile, the reliability and the real-time performance of fault diagnosis are improved, and powerful support is provided for intelligent maintenance of an automobile axle.

Description

Technical Field

[0001] This application relates to the technical field of fault diagnosis, and more specifically, to a method and system for diagnosing automobile axle faults. Background Art

[0002] As a core component for power transmission and load bearing of a vehicle, the health status of an automobile axle is directly related to the safety, stability, and service life of the vehicle. During the long-term operation of the vehicle, the axle may generate abnormal noises due to factors such as mechanical wear, assembly errors, material fatigue, or external impacts. These abnormal noises are often typical characteristics of early faults. However, traditional axle fault diagnosis methods have limitations such as relying on manual experience, high detection costs, and insufficient real-time performance, making it difficult to meet the intelligent fault diagnosis requirements of modern automobiles.

[0003] In recent years, fault diagnosis technology based on acoustic signals has received attention due to its advantages such as non-contact acquisition and rich information. For example, in the field of rotating machinery, the operating noise of equipment is collected through a microphone array, and machine learning algorithms are combined to achieve fault classification of components such as bearings and gears. However, due to the characteristics of axle abnormal noise signals being non-stationary and having a low signal-to-noise ratio, and being highly coupled with vehicle operating environment noises (such as tire noise and wind noise), relying solely on the processing of single-modal sound signals may make it difficult to effectively distinguish fault-related acoustic patterns, resulting in limited accuracy of fault diagnosis.

[0004] Therefore, an optimized method and system for diagnosing automobile axle faults are expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a method and system for diagnosing automobile axle faults. The method collects axle abnormal noise sound signals during driving through a sound sensor, uses a data processing algorithm based on deep learning to perform fluctuation analysis on the axle abnormal noise sound signals, and combines the abnormal noise description information input by the user to strengthen the understanding of the characteristics of the axle abnormal noise sound. Through fine-grained cross-modal response collaborative reasoning on the fluctuation patterns of the axle abnormal noise sound signals and the abnormal noise description information, potential axle fault characteristics are excavated to achieve intelligent diagnosis of automobile axle fault components. In this way, the dependence on manual experience in traditional automobile axle fault diagnosis methods can be effectively overcome, while improving the reliability and real-time performance of fault diagnosis, providing strong support for the intelligent maintenance of automobile axles.

[0006] According to one aspect of this application, a method for diagnosing automobile axle faults is provided, which includes: Obtaining axle abnormal noise sound signals collected by a sound sensor; Obtaining abnormal noise description information input by the user; Extract the sound fluctuation characteristics from the axle abnormal sound signal to obtain the axle abnormal sound waveform characteristics; Perform semantic embedding encoding on the abnormal sound description information to obtain the semantic encoding characteristics of the abnormal sound description information; Perform cross-modal response collaborative reasoning based on granular interaction iteration on the axle abnormal sound waveform characteristics and the semantic encoding characteristics of the abnormal sound description information to obtain the abnormal sound feature - abnormal sound description fine-grained interaction response inference encoding characteristics; Based on the abnormal sound feature - abnormal sound description fine-grained interaction response inference encoding characteristics, determine the faulty component of the vehicle axle.

[0007] According to another aspect of the present application, there is provided a vehicle axle fault diagnosis system, which includes: An abnormal sound signal acquisition module for acquiring the axle abnormal sound signal collected by a sound sensor; An abnormal sound description information acquisition module for acquiring the abnormal sound description information input by a user; A sound fluctuation characteristic extraction module for extracting the sound fluctuation characteristics from the axle abnormal sound signal to obtain the axle abnormal sound waveform characteristics; A semantic embedding encoding module for performing semantic embedding encoding on the abnormal sound description information to obtain the semantic encoding characteristics of the abnormal sound description information; A cross-modal response collaborative reasoning module for performing cross-modal response collaborative reasoning based on granular interaction iteration on the axle abnormal sound waveform characteristics and the semantic encoding characteristics of the abnormal sound description information to obtain the abnormal sound feature - abnormal sound description fine-grained interaction response inference encoding characteristics; A fault diagnosis module for determining the faulty component of the vehicle axle based on the abnormal sound feature - abnormal sound description fine-grained interaction response inference encoding characteristics.

[0008] Compared with the prior art, the vehicle axle fault diagnosis method and system provided by the present application collect the axle abnormal sound signal during driving through a sound sensor, perform fluctuation analysis on the axle abnormal sound signal by using a data processing algorithm based on deep learning, and at the same time combine the abnormal sound description information input by the user to strengthen the understanding of the axle abnormal sound characteristics. By performing fine-grained cross-modal response collaborative reasoning on the fluctuation pattern of the axle abnormal sound signal and the abnormal sound description information, potential axle fault characteristics are excavated to realize the intelligent diagnosis of the faulty component of the vehicle axle. In this way, the dependence on manual experience in traditional vehicle axle fault diagnosis methods can be effectively overcome, and at the same time, the reliability and real-time performance of fault diagnosis are improved, providing strong support for the intelligent maintenance of vehicle axles. Brief Description of the Drawings

[0009] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of an automobile axle fault diagnosis method according to an embodiment of the present application.

[0011] Figure 2 It is a schematic diagram of data flow of an automobile axle fault diagnosis method according to an embodiment of the present application.

[0012] Figure 3 It is a flowchart of sub-step S5 of an automobile axle fault diagnosis method according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of sub-step S52 of an automobile axle fault diagnosis method according to an embodiment of the present application.

[0014] Figure 5 It is a block diagram of an automobile axle fault diagnosis system according to an embodiment of the present application. Detailed implementation manners

[0015] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "including" and "comprising" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0016] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0017] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in order. On the contrary, as needed, various steps can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0018] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0019] To address the technical problems described in the above background art, the present application proposes a method for diagnosing faults in an automotive axle. It collects the abnormal sound signals of the axle during driving through a sound sensor, and uses a data processing algorithm based on deep learning to perform fluctuation analysis on the abnormal sound signals of the axle. At the same time, it combines the abnormal sound description information input by the user to strengthen the understanding of the characteristics of the abnormal sound of the axle. Through fine-grained cross-modal response collaborative reasoning on the fluctuation patterns of the abnormal sound signals of the axle and the abnormal sound description information, it excavates potential fault characteristics of the axle to achieve intelligent diagnosis of the faulty components of the automotive axle. In this way, it can effectively overcome the dependence on manual experience in traditional automotive axle fault diagnosis methods, and at the same time improve the reliability and real-time performance of fault diagnosis, providing strong support for the intelligent maintenance of automotive axles.

[0020] Figure 1 It is a flowchart of the method for diagnosing faults in an automotive axle according to an embodiment of the present application. Figure 2 It is a schematic diagram of the data flow of the method for diagnosing faults in an automotive axle according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the method for diagnosing faults in an automotive axle includes the steps of: S1, obtaining the abnormal sound signals of the axle collected by the sound sensor; S2, obtaining the abnormal sound description information input by the user; S3, extracting the sound fluctuation characteristics from the abnormal sound signals of the axle to obtain the waveform characteristics of the abnormal sound of the axle; S4, performing semantic embedding encoding on the abnormal sound description information to obtain the semantic encoding characteristics of the abnormal sound description information; S5, performing fine-grained cross-modal response collaborative reasoning based on granularity interaction iteration on the waveform characteristics of the abnormal sound of the axle and the semantic encoding characteristics of the abnormal sound description information to obtain the fine-grained interaction response inference encoding characteristics of the abnormal sound characteristics - abnormal sound description; S6, based on the fine-grained interaction response inference encoding characteristics of the abnormal sound characteristics - abnormal sound description, determining the faulty components of the automotive axle.

[0021] In the above vehicle axle fault diagnosis method, in step S1, an abnormal sound signal of the vehicle axle collected by a sound sensor is obtained. It should be understood that as a key component of the vehicle running system, when a vehicle axle fails, abnormal movement of its internal mechanical structure will cause air vibration and generate abnormal sounds. Based on the fault diagnosis technology of acoustic signals, with its non-contact acquisition characteristics, it will not interfere with the normal operation of the vehicle axle, and at the same time can monitor the working state of the vehicle axle in real time, becoming an important way for vehicle axle fault diagnosis. Based on this, in this application, by using a non-contact sound sensor (such as a high-sensitivity microphone or an acoustic array) to collect the abnormal sound signal of the vehicle axle, mechanical contact interference can be avoided, and at the same time, wide-band acoustic wave signals containing rich fault information can be captured, providing a raw data basis for subsequent analysis.

[0022] In the specific implementation process, it is necessary to install a high-sensitivity microphone or an acoustic array at a position close to the vehicle axle at the bottom of the vehicle to ensure that the sound signals emitted from the vehicle axle can be comprehensively captured. The selection of these positions is based on an in-depth understanding of the working environment of the vehicle axle. That is, as a core component for power transmission and load bearing, the vehicle axle may generate abnormal sounds due to factors such as mechanical wear, assembly errors, material fatigue, or external impacts during long-term operation. By reasonably arranging the sound sensors, external interference can be effectively reduced and the quality of the collected data can be improved. In addition, considering the space limitation and complex operating environment around the vehicle axle, appropriate protection measures for the sound sensors are required, such as waterproof and dustproof designs, to ensure that the sensors can work stably under various harsh conditions.

[0023] In order to accurately capture wide-band acoustic wave signals, the frequency response range of the sensor needs to be carefully selected. The abnormal sound of the vehicle axle often contains information of various different frequency components, which means that the sensor used should have a wide frequency response ability to capture as many acoustic patterns related to faults as possible. At the same time, the dynamic range of the sensor is also crucial because the intensity of the abnormal sound signal of the vehicle axle may vary greatly. A good dynamic range helps to handle various situations from slight background noise to strong abnormal sounds without losing details.

[0024] In addition to hardware considerations, data synchronization and calibration at the software level cannot be ignored. Since the axle abnormal noise sound signal is the basis for subsequent analysis, it is necessary to ensure that the collected data has a high degree of temporal consistency and accuracy. For this purpose, using GPS time synchronization technology or an internal high-precision clock can provide an accurate timestamp for each sampling, enabling data from different sources to be compared and analyzed on the same time axis. Further, considering that the axle abnormal noise signal is usually closely related to vehicle operating environment noise (such as tire noise, wind noise), it is particularly important to take effective noise reduction measures. On the one hand, the impact of environmental noise can be naturally reduced by optimizing the sensor layout; on the other hand, by means of signal processing technologies such as digital filters, unnecessary noise components can be removed in the later processing stage to highlight the fault characteristics.

[0025] In the process of specific implementation, it is also necessary to consider how to effectively manage a large amount of continuously collected sound data. The amount of sound signals generated by the axle during long-term operation is huge, and directly storing all unfiltered data is neither economical nor efficient. For this purpose, certain trigger conditions can be preset, such as automatically starting the recording function when a sudden increase in sound intensity within a specific frequency range is detected, or dynamically adjusting the sampling rate according to changes in driving parameters such as vehicle speed and acceleration, so as to reduce unnecessary data accumulation while ensuring that key fault information is not missed.

[0026] In the above method for diagnosing automobile axle faults, in step S2, obtain the abnormal noise description information input by the user. It should be understood that in this application, considering the axle fault diagnosis scenario, the fault analysis relying solely on the sound signal may be interfered by environmental noise, resulting in inaccurate diagnosis results. As the direct operator of the vehicle, the user can obtain axle abnormal noise information from the intuitive feeling level during daily driving. For example, the specific operating conditions of the vehicle at the moment when the abnormal noise appears (such as sudden braking, sudden acceleration, during gear shifting), the duration of the abnormal noise, and the subjective feeling of the sound (sharp and harsh, dull and low, periodic clicking sound, etc.). Therefore, this application further obtains the user's subjective description of the axle abnormal noise (such as "intermittent clicking sound from the right rear wheel when turning at low speed"), and based on the fault semantic features (such as location, frequency characteristics, trigger conditions) contained in the description, to make up for the lack of implicit semantic information in the pure physical signal, thereby enhancing the fine-grained understanding ability of complex axle fault modes.

[0027] In the specific implementation process, through the in-vehicle information system or a specially developed mobile application, users can be guided to provide as detailed a description of abnormal noises as possible. On the interface, a series of open-ended questions are set to encourage users to share all relevant experiences when they hear abnormal noises. These questions may include the operations being performed when the abnormal noise occurs (such as hard braking, sudden acceleration, gear shifting), the duration and frequency characteristics of the abnormal noise (whether it is continuous or intermittent), and the subjective perception of the sound (whether the sound is high-pitched or low-pitched). Such an open-ended questioning method allows users to freely express according to their memories, thereby capturing implicit semantic information that is difficult to obtain through traditional physical signal analysis.

[0028] Furthermore, to enhance the user experience, some example descriptions or prompts can be provided before the user submits the description to help the user better understand and answer the questions. For example, show some typical abnormal noise cases and their corresponding descriptions to let the user know what kind of information is valuable. This approach can not only lower the participation threshold for users but also ensure that the collected data has high quality and consistency. In addition, customize and adjust the language style and questioning method of the interface according to the characteristics of different user groups to meet the needs of a wider range of users.

[0029] In addition to collecting abnormal noise description information through the in-vehicle system or mobile application, social media platforms can also be considered as information collection channels. Many car owners will share the vehicle problems they encounter on social networks, and these public information also contains rich fault clues. By developing an intelligent crawler program, posts and comments related to axle abnormal noises can be automatically captured and analyzed to extract valuable fault descriptions. It should be noted that although the abnormal noise description information provided by users is extremely valuable, its accuracy and reliability are still affected by individual experiences and perceptual differences. Therefore, combine data from other sources, such as sound signals and vehicle operation parameters collected by sensors, for comprehensive analysis to maximize the role of user descriptions.

[0030] In the above-mentioned automobile axle fault diagnosis method, the step S3 extracts the sound fluctuation characteristics from the abnormal sound signal of the axle to obtain the abnormal sound waveform characteristics of the axle. In a specific example of the present application, the step S3 includes: extracting the sound fluctuation characteristics of the abnormal sound signal of the axle based on one-dimensional convolution coding to obtain the abnormal sound waveform characteristic coding vector of the axle as the abnormal sound waveform characteristic of the axle. It should be understood that since the abnormal sound signal of the axle has non-stationary characteristics (such as impact transients and modulation sidebands), traditional spectrum analysis (FFT) is difficult to capture its time-varying details. Therefore, the present application adopts a one-dimensional convolutional neural network (1D-CNN) to extract the characteristics of the abnormal sound signal of the axle. Specifically, 1D-CNN has unique advantages in processing one-dimensional time series signals. Based on the local receptive field and weight sharing mechanism, it can automatically learn the local features and fluctuation patterns of sound signals in the time dimension (such as transient pulses, resonant frequency shifts) through layer-by-layer convolution and feature extraction, and encode them into high-discrimination axle abnormal sound waveform feature encoding vectors, thereby providing effective data representation for subsequent fault analysis and diagnosis.

[0031] In the above-mentioned automobile axle fault diagnosis method, the step S4 performs semantic embedding encoding on the abnormal sound description information to obtain the semantic encoding features of the abnormal sound description information. In a specific example of the present application, the step S4 includes: performing semantic embedding encoding on the abnormal sound description information based on the Transformer model to obtain the semantic encoding vector of the abnormal sound description information as the semantic encoding features of the abnormal sound description information. It should be understood that in order to realize the interactive analysis between the abnormal sound description information input by the user and the fluctuation characteristics of the abnormal sound signal of the axle, it is necessary to further convert the abnormal sound description information into a semantic vector representation aligned with the acoustic feature space. To this end, the present application adopts the Transformer model to perform semantic embedding encoding on the abnormal sound description information. Those of ordinary skill in the art should know that the Transformer model, with its self-attention mechanism and position encoding capabilities, can effectively capture text context dependencies (such as the position association between "low-speed turning" and "right rear wheel") and generate a high-dimensional embedding representation containing text context semantics. Specifically, first, the abnormal sound description information input is initialized using pre-trained word vectors, and each word in the abnormal sound description information is mapped to a continuous vector space to obtain an initial word vector representation. Subsequently, sinusoidal position encoding is added to each initial word vector representation of the abnormal sound description information text, and the relative position of each word in the text is marked. Then, through a multi-layer Transformer encoding structure, the long-distance dependency between words is captured based on the self-attention mechanism, and high-level contextual semantic features are gradually abstracted. The semantic encoding vector of the abnormal sound description information is output, providing a fine-grained semantic understanding basis for subsequent axle fault analysis.

[0032] In the above vehicle axle fault diagnosis method, in step S5, a cross-modal response collaborative reasoning based on granular interaction iteration is performed on the waveform feature of the axle abnormal sound and the semantic coding feature of the abnormal sound description information to obtain an abnormal sound feature - abnormal sound description fine-grained interaction response inference coding feature. Specifically, since the axle abnormal sound signal and the abnormal sound description information come from the physical layer and the semantic layer respectively, there is heterogeneity in their feature spaces. Therefore, in order to fully explore the potential correlation between the axle abnormal sound signal and the abnormal sound description information and achieve fine-grained feature interaction, the present application proposes a cross-modal response collaborative reasoning method based on granular interaction iteration. By performing fine-grained local interaction response coding on the waveform feature coding vector of the axle abnormal sound and the semantic coding vector of the abnormal sound description information, a fine-grained correlation relationship between features is constructed (for example, the correspondence between the "clunking sound" described by the user and the transient pulse peak in the acoustic signal), and based on the local feature interaction states of each, feature attention modulation and chained iterative reasoning are performed to gradually update the feature representation to approximate the common potential fault feature space, strengthen the fusion of relevant features, suppress the interference of irrelevant features, and finally achieve dynamic complementarity and collaborative enhancement of cross-modal features, and output a global abnormal sound feature - abnormal sound description fine-grained interaction response inference coding vector. Among them, Figure 3 is a flowchart of sub-step S5 of the vehicle axle fault diagnosis method according to an embodiment of the present application. As Figure 3 shown, step S5 includes the steps of: S51, performing local granular feature interaction response coding on the waveform feature coding vector of the axle abnormal sound and the semantic coding vector of the abnormal sound description information to obtain a set of abnormal sound fluctuation - abnormal sound description local implicit feature interaction response coding vectors; S52, performing feature distribution significance modulation on the set of abnormal sound fluctuation - abnormal sound description local implicit feature interaction response coding vectors to obtain a set of modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response coding vectors; S53, performing iterative reasoning on the set of modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response coding vectors to obtain an abnormal sound feature - abnormal sound description fine-grained interaction response inference coding vector as the abnormal sound feature - abnormal sound description fine-grained interaction response inference coding feature.

[0033] Specifically, step S51 includes: First, performing local implicit feature extraction based on one-dimensional convolutional coding on the waveform feature coding vector of the axle abnormal sound and the semantic coding vector of the abnormal sound description information respectively to obtain a set of local implicit feature coding vectors of the axle abnormal sound waveform and a set of local implicit feature coding vectors of the abnormal sound description information semantics, which is expressed by the formula: ; where Represents the encoding vector of the sound waveform feature of the axle abnormal noise, Represents the one-dimensional convolution encoding with a convolution kernel scale of, and Respectively represent the 1st, 2nd, th, th, and th local implicit feature encoding vectors in the set of local implicit feature encoding vectors of the axle abnormal noise sound waveform, Represents the semantic encoding vector of the abnormal noise description information, and Respectively represent the 1st, 2nd, th, th, and

[0034] That is, through the design of the local receptive field of one-dimensional convolution, it can effectively capture the local fluctuation patterns of the time series of the sound signal and the local context correlation of the text semantics, thereby converting the encoding vector of the axle abnormal noise sound waveform feature and the semantic encoding vector of the abnormal noise description information into a set of local implicit feature encoding vectors of the axle abnormal noise sound waveform with physical interpretability and a set of local implicit feature encoding vectors of the abnormal noise description information semantics. Through this one-dimensional convolution method, it provides a feature expression with spatio-temporal context sensitivity for subsequent cross-modal reasoning, enabling a fine-grained correspondence between the short-time frequency domain features in the sound signal and the precise semantic fragments in the text description, and ultimately enhancing the identification ability of weak fault features in a coupled noise environment.

[0035] Then, for each group of corresponding local implicit feature encoding vectors of the axle abnormal noise sound waveform and local implicit feature encoding vectors of the abnormal noise description information semantics in the set of local implicit feature encoding vectors of the axle abnormal noise sound waveform and the set of local implicit feature encoding vectors of the abnormal noise description information semantics, perform single-entity feature interaction respectively to obtain the set of local implicit feature interaction response encoding vectors of the abnormal noise fluctuation - abnormal noise description, which is expressed by the formula: ; Among them, Represents the concatenation function, Represents the dot product operation, Represents the dot addition operation, Represents the dot subtraction operation, Represents the bias term, Represents the weight matrix, Represents the An abnormal sound fluctuation - local implicit feature interaction response coding vector of abnormal sound description.

[0036] That is, through single - feature interaction, for each corresponding coding vector of local implicit features of the abnormal sound waveform of the axle and the coding vector of local implicit features of the semantic information of the abnormal sound description, the co - activation or inhibition relationship between cross - modal features can be dynamically learned, and a physically interpretable mapping between cross - modal features can be established at a fine - grained level. The short - term abnormal patterns in the sound signal and the key semantic clues in the user description are jointly characterized, thereby suppressing the pollution of single - modal noise to feature expression. The set of generated abnormal sound fluctuation - local implicit feature interaction response coding vectors of abnormal sound description can significantly enhance the model's robust expression ability for weak fault features in a coupled noise environment.

[0037] Figure 4 It is a flowchart of sub - step S52 of the vehicle axle fault diagnosis method according to an embodiment of the present application. As Figure 4 shown, the step S52 includes steps: S521, based on the feature distribution characteristics of each abnormal sound fluctuation - local implicit feature interaction response coding vector in the set of abnormal sound fluctuation - local implicit feature interaction response coding vectors, determining the chain - reasoning attention weights of each abnormal sound fluctuation - local implicit feature interaction response coding vector to obtain a set of abnormal sound fluctuation - abnormal sound description chain - reasoning attention weights; S522, based on the set of abnormal sound fluctuation - abnormal sound description chain - reasoning attention weights, performing feature modulation on the set of abnormal sound fluctuation - local implicit feature interaction response coding vectors to obtain the set of modulated abnormal sound fluctuation - local implicit feature interaction response coding vectors.

[0038] More specifically, the step S521 is expressed by the formula: ; where represents the normalized exponential function, represents 's feature scale value, represents calculating the norm, represents the th feature value at the th position in the th abnormal sound fluctuation - local implicit feature interaction response coding vector, represents the abnormal sound fluctuation - abnormal sound description chain - reasoning attention weight of

[0039] That is, through the calculation of chain reasoning attention weights, cross-modal interaction segments strongly related to the physical faults of the axle can be screened based on the statistical characteristics of local interaction responses, and a hierarchical reasoning attention focusing mechanism can be constructed to preferentially activate cross-modal interaction features with high confidence and suppress the interference of redundant or low-correlation segments on subsequent reasoning. At the same time, through the generation of the set of attention weights, the dynamic optimization of the fault feature reasoning path can be realized, improving the feature screening efficiency of the model in complex noise scenarios, providing an interpretable weight assignment basis for chain reasoning, and significantly reducing the misjudgment probability caused by accidental noise.

[0040] In a preferred example of the present application, the step S522 includes: First, based on the decomposition of the interaction space effect of each abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vector in the set of abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors, the set of abnormal sound fluctuation-abnormal sound description chain reasoning attention weights is compensated and corrected to obtain the corrected set of abnormal sound fluctuation-abnormal sound description chain reasoning attention weights, which is expressed by the formula: ; Wherein, and respectively represent and The statistical quantity of the feature interaction translation action vector between, represents and The statistical quantity of the feature interaction fluctuation action vector between, represents the natural logarithm with base e, represents the periodic local rule compensation factor, represents the exponential function with base e, represents the periodic local auxiliary phase change factor, and respectively represent different weighting values, represents The corresponding corrected abnormal sound fluctuation-abnormal sound description chain reasoning attention weight.

[0041] That is, by decomposing the interaction spatial effect between the local implicit feature encoding vector of the axle abnormal sound waveform and the local implicit feature encoding vector of the semantic of the abnormal sound description information, the translational effect and the fluctuation effect of the feature interaction are decoupled, so as to quantify the contribution differences of different interaction modes to the attention weight. Specifically, the present application improves the correlation between different interaction modes in the abnormal sound fluctuation - abnormal sound description interaction space by introducing the direction sensitivity of the translational effect and the periodic characteristics of the fluctuation effect, so that the corrected abnormal sound fluctuation - abnormal sound description chain inference attention weight can more accurately capture the spatial action mode strongly related to physical faults in cross-modal interaction. Finally, through the compensation and correction mechanism, the adaptability of the corrected abnormal sound fluctuation - abnormal sound description chain inference attention weight to complex fault scenarios can be improved, the false detection rate caused by the confusion of different interaction space action modes can be effectively suppressed, and weight guidance conforming to the physical laws of mechanical faults can be provided for subsequent chain inference.

[0042] Then, based on the set of the corrected abnormal sound fluctuation - abnormal sound description chain inference attention weights, the set of the abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors is weighted and modulated to obtain the set of the modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors, which is expressed by the formula: ; Wherein, and respectively represent the first, the second, the th, and the th modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors in the set of the modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors.

[0043] That is, through the dynamic scaling of the weight-driven feature space, the high-weight interaction vectors dominate the feature expression, so as to realize the controllable enhancement and suppression of cross-modal interaction features, make the subsequent inference process focus on the key interaction modes strongly related to the physical faults of the axle, and at the same time weaken the influence of low-confidence interaction segments caused by environmental noise or user subjective description deviation. The finally generated modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors effectively reduce redundant information and can provide high-purity and strongly discriminative cross-modal feature inputs for the classification of subsequent faulty components.

[0044] Specifically, in a specific example of the present application, the step S53 includes: performing iterative inference on the set of the modulated abnormal sound fluctuation - abnormal sound description local implicit feature interaction response encoding vectors based on a forward LSTM model to obtain the abnormal sound feature - abnormal sound description fine-grained interaction response inference encoding vectors, which is expressed by the formula: ; Among them, represents the forward LSTM model, represents the abnormal sound feature - abnormal sound description fine-grained interaction response inference coding vector.

[0045] It should be understood that the LSTM (Long Short-Term Memory) model has the ability to capture long-distance dependencies in sequential data. In this application, through the iterative inference and gating mechanism of the forward LSTM model, the context dependence between the abnormal sound wave - abnormal sound description local interaction features can be deeply mined, so that the generated abnormal sound feature - abnormal sound description fine-grained interaction response inference coding vector can better reflect the evolution law and potential correlation of the axle physical fault, improve the capture ability of time-series sensitive features in fault diagnosis, and provide an interpretable inference path that conforms to the development law of mechanical faults for fault diagnosis.

[0046] In the above method for diagnosing automobile axle faults, in step S6, based on the inference coding features of the abnormal sound feature-abnormal sound description fine-grained interaction response, the faulty component of the automobile axle is determined. In a specific example of the present application, step S6 includes: inputting the inference coding vector of the abnormal sound feature-abnormal sound description fine-grained interaction response into a fault diagnosis module based on a classifier to obtain the recognition result of the faulty component of the automobile axle. That is, inputting the inference coding vector of the abnormal sound feature-abnormal sound description fine-grained interaction response into a pre-trained classifier, which is constructed based on a deep learning framework and trained with a large number of labeled axle fault samples to learn the mapping relationship from the acoustic characteristics of axle abnormal sounds to faulty components. After receiving the inference coding vector of the abnormal sound feature-abnormal sound description fine-grained interaction response, the classifier performs feature analysis and classification judgment based on the prior knowledge learned internally, and based on the learned fluctuation characteristics and semantic description information of the abnormal sounds of the automobile axle, identifies the faulty components of the automobile axle, such as "bearing damage", "gear wear", or "drive shaft fracture", etc., thereby realizing the intelligent diagnosis of the faulty components of the automobile axle. Specifically, taking the multi-layer perceptron (MLP) in the neural network classifier as an example, in the training stage, a large amount of data of different axle fault cases are collected, including axle abnormal sound signals and user description information. After being processed by the foregoing steps, the fine-grained interaction response inference coding features between the two are obtained, and at the same time, the corresponding known faulty component labels are combined. These data are divided into a training set, a validation set, and a test set according to a certain ratio. The training set data is input into the MLP. The MLP performs linear transformation and non-linear activation on the input features through multiple fully connected layers, and continuously adjusts the weights and biases of the fully connected layers through the backpropagation algorithm to minimize the error between the predicted faulty component category of the model and the true faulty component label. After multiple rounds of training, the model gradually learns the feature patterns corresponding to different faulty components. In the diagnosis stage, the inference coding vector of the abnormal sound feature-abnormal sound description fine-grained interaction response to be diagnosed is input into the trained MLP. The MLP calculates and judges according to the learned feature patterns and outputs the most likely faulty component category. In this way, not only the reliability and efficiency of fault diagnosis are improved, but also the dependence on manual experience is reduced, providing a reliable basis for automobile repair and maintenance.

[0047] In summary, the vehicle axle fault diagnosis method based on the embodiments of the present application is elucidated. It collects the abnormal sound signals of the vehicle axle during driving through a sound sensor, and uses a data processing algorithm based on deep learning to perform fluctuation analysis on the abnormal sound signals of the vehicle axle. At the same time, it combines the abnormal sound description information input by the user to strengthen the understanding of the characteristics of the abnormal sound of the vehicle axle. Through fine-grained cross-modal response collaborative reasoning on the fluctuation mode of the abnormal sound signals of the vehicle axle and the abnormal sound description information, potential vehicle axle fault characteristics are excavated to achieve intelligent diagnosis of the faulty components of the vehicle axle. In this way, the dependence on manual experience in traditional vehicle axle fault diagnosis methods can be effectively overcome, while the reliability and real-time performance of fault diagnosis are improved, providing strong support for the intelligent maintenance of vehicle axles.

[0048] Furthermore, a vehicle axle fault diagnosis system is also provided.

[0049] Figure 5 The block diagram of the vehicle axle fault diagnosis system according to the embodiments of the present application is shown in FIG. Figure 5 As shown, the vehicle axle fault diagnosis system 100 according to the embodiments of the present application includes: an abnormal sound signal acquisition module 110, configured to obtain the abnormal sound signals of the vehicle axle collected by the sound sensor; an abnormal sound description information acquisition module 120, configured to obtain the abnormal sound description information input by the user; a sound fluctuation feature extraction module 130, configured to extract sound fluctuation features from the abnormal sound signals of the vehicle axle to obtain the waveform features of the abnormal sound of the vehicle axle; a semantic embedding encoding module 140, configured to perform semantic embedding encoding on the abnormal sound description information to obtain the semantic encoding features of the abnormal sound description information; a cross-modal response collaborative reasoning module 150, configured to perform cross-modal response collaborative reasoning based on granulated interaction iteration on the waveform features of the abnormal sound of the vehicle axle and the semantic encoding features of the abnormal sound description information to obtain the inference encoding features of the fine-grained interaction response of the abnormal sound features - abnormal sound description; and a fault diagnosis module 160, configured to determine the faulty components of the vehicle axle based on the inference encoding features of the fine-grained interaction response of the abnormal sound features - abnormal sound description.

[0050] Here, those skilled in the art can understand that the specific operations of each module in the above vehicle axle fault diagnosis system have been introduced in detail in the description of the vehicle axle fault diagnosis method above, and therefore, the repeated description thereof will be omitted. Figures 1 to 4 of the vehicle axle fault diagnosis method and thus, the repeated description thereof will be omitted.

[0051] The basic principles of the present invention have been described above in connection with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. Additionally, the specific details of the above embodiments are only for the purpose of illustration and facilitating understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details for implementation.

[0052] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there can be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0053] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes within the meaning and scope of the equivalent elements of the claims in the present invention. Any associated drawing reference signs in the claims should not be regarded as limiting the claimed rights.

[0054] In addition, obviously the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0055] Finally, it should be noted that the above description has been given for the purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for diagnosing vehicle axle faults, characterized in that: include: Acquiring abnormal sound signals of the axle collected by the sound sensor; Obtain abnormal sound description information input by the user; Extracting sound fluctuation characteristics from the axle abnormal sound signal to obtain axle abnormal sound waveform characteristics; Performing semantic embedding coding on the abnormal sound description information to obtain semantic coding features of the abnormal sound description information; Performing cross-modal response collaborative reasoning based on granular interactive iteration on the abnormal sound waveform feature of the axle and the semantic coding feature of the abnormal sound description information to obtain abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding feature; Based on the abnormal sound feature-abnormal sound description fine-grained interactive response inference coding feature, the faulty component of the automobile axle is determined.

2. The automobile axle fault diagnosis method according to claim 1, characterized in that: Extracting sound fluctuation features from the axle abnormal sound signal to obtain axle abnormal sound waveform features includes: The sound fluctuation feature of the abnormal sound of the vehicle bridge is extracted based on one-dimensional convolution coding to obtain a waveform feature coding vector of the abnormal sound of the vehicle bridge as the waveform feature of the abnormal sound of the vehicle bridge.

3. The automobile axle fault diagnosis method according to claim 2, characterized in that: The abnormal sound description information is semantically embedded and encoded to obtain semantic encoding features of the abnormal sound description information, including: The abnormal sound description information is semantically embedded and encoded based on a Transformer model to obtain a semantic encoding vector of the abnormal sound description information as a semantic encoding feature of the abnormal sound description information.

4. The automobile axle fault diagnosis method according to claim 3, characterized in that: The abnormal sound waveform feature of the axle and the semantic coding feature of the abnormal sound description information are subjected to cross-modal response collaborative reasoning based on granular interactive iteration to obtain abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding features, including: Performing local granular feature interactive response coding on the axle abnormal sound waveform feature coding vector and the abnormal sound description information semantic coding vector to obtain a set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors; Performing feature distribution significance modulation on the set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors to obtain a set of modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors; Iterative reasoning is performed on the set of the modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors to obtain the abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding vector as the abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding feature.

5. The automobile axle fault diagnosis method according to claim 4, characterized in that: The axle abnormal sound waveform feature coding vector and the abnormal sound description information semantic coding vector are subjected to local granular feature interactive response coding to obtain a set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors, including: Performing local implicit feature extraction based on one-dimensional convolution coding on the axle abnormal sound waveform feature coding vector and the abnormal sound description information semantic coding vector respectively to obtain a set of axle abnormal sound waveform local implicit feature coding vectors and a set of abnormal sound description information semantic local implicit feature coding vectors; Perform monomer feature interaction on each corresponding group of local implicit feature coding vectors of the abnormal sound waveform of the axle and the local implicit feature coding vectors of the semantics of the abnormal sound description information in the set of local implicit feature coding vectors of the abnormal sound waveform of the axle and the set of local implicit feature coding vectors of the semantics of the abnormal sound description information to obtain the set of the abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors.

6. The automobile axle fault diagnosis method according to claim 5, characterized in that: The set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors is subjected to feature distribution significance modulation to obtain a set of modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors, including: Based on the characteristic distribution characteristics of each abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vector in the set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors, determine the chain reasoning attention weights of each abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vector to obtain a set of abnormal sound fluctuation-abnormal sound description chain reasoning attention weights; Based on the set of abnormal sound fluctuation-abnormal sound description chain reasoning attention weights, feature modulation is performed on the set of abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors to obtain the set of modulated abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors.

7. The automobile axle fault diagnosis method according to claim 6, characterized in that: Based on the set of abnormal sound fluctuation-abnormal sound description chained reasoning attention weights, feature modulation is performed on the set of abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors to obtain the set of modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors, including: Based on the interaction space action decomposition of each abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vector in the set of the abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vector, the set of abnormal sound fluctuation-abnormal sound description chain reasoning attention weights is compensated and corrected to obtain a set of corrected abnormal sound fluctuation-abnormal sound description chain reasoning attention weights; Based on the set of modified abnormal sound fluctuation-abnormal sound description chain reasoning attention weights, the set of abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors is weighted modulated to obtain the set of modulated abnormal sound fluctuation-abnormal sound description local implicit feature interaction response coding vectors.

8. The automobile axle fault diagnosis method according to claim 7, characterized in that: Iteratively reasoning the set of the modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors to obtain the abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding vector, including: The set of the modulated abnormal sound fluctuation-abnormal sound description local implicit feature interactive response coding vectors is iteratively inferred based on the forward LSTM model to obtain the abnormal sound feature-abnormal sound description fine-grained interactive response inference coding vector.

9. The automobile axle fault diagnosis method according to claim 8, characterized in that: Based on the abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding feature, the faulty component of the automobile axle is determined, including: The abnormal sound feature-abnormal sound description fine-grained interactive response inference encoding vector is input into a classifier-based fault diagnosis module to obtain the identification result of the automobile axle fault component.

10. An automobile axle fault diagnosis system, characterized in that: include: The abnormal sound signal acquisition module is used to acquire the abnormal sound signal of the axle collected by the sound sensor; The abnormal sound description information acquisition module is used to obtain the abnormal sound description information input by the user; A sound fluctuation feature extraction module, used for extracting the sound fluctuation feature from the axle abnormal sound signal to obtain the axle abnormal sound waveform feature; A semantic embedding coding module, used for performing semantic embedding coding on the abnormal sound description information to obtain semantic coding features of the abnormal sound description information; A cross-modal response collaborative reasoning module is used to perform cross-modal response collaborative reasoning based on granular interactive iteration on the abnormal sound waveform feature of the axle and the semantic coding feature of the abnormal sound description information to obtain the abnormal sound feature-abnormal sound description fine-grained interactive response reasoning coding feature; The fault diagnosis module is used to determine the faulty component of the automobile axle based on the abnormal sound feature-abnormal sound description fine-grained interactive response inference coding feature.

Images